xudong-yan / scadn Goto Github PK

博主您好，我在train的时候，遇见些问题，他报错：
Traceback (most recent call last):
File "main.py", line 146, in
main()
File "main.py", line 51, in main
model.train()
File "/root/autodl-tmp/project/src/experiments.py", line 76, in train
self.inpaint_model.backward(gen_loss, dis_loss)
File "/root/autodl-tmp/project/src/models.py", line 175, in backward
gen_loss.backward()
File "/root/miniconda3/lib/python3.8/site-packages/torch/_tensor.py", line 363, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "/root/miniconda3/lib/python3.8/site-packages/torch/autograd/init.py", line 173, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 512, 4, 4]] is at version 3; expected version 2 instead.
我采用异常跟踪，给我报：
/root/miniconda3/lib/python3.8/site-packages/torch/autograd/init.py:173: UserWarning: Error detected in DivBackward0. Traceback of forward call that caused the error:
File "main.py", line 146, in
main()
File "main.py", line 51, in main
model.train()
File "/root/autodl-tmp/project/src/experiments.py", line 73, in train
outputs, gen_loss, dis_loss, logs = self.inpaint_model.process(images, masks)
File "/root/autodl-tmp/project/src/models.py", line 124, in process
gen_fake, gen_fake_feat = self.discriminator(gen_input_fake) # in: [rgb(3)]
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/root/autodl-tmp/project/src/networks.py", line 137, in forward
conv5 = self.conv5(conv4)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/container.py", line 141, in forward
input = module(input)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1117, in _call_impl
result = hook(self, input)
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/utils/spectral_norm.py", line 105, in call
setattr(module, self.name, self.compute_weight(module, do_power_iteration=module.training))
File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/utils/spectral_norm.py", line 92, in compute_weight
weight = weight / sigma
(Triggered internally at ../torch/csrc/autograd/python_anomaly_mode.cpp:104.)
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
好像是spectral_norm中的除法出现问题，不知道该如何修改，或者会不会是pytorch版本问题，还请博主帮我解答

Is there any code for the whole process of training? It would be appreciated if you could provide it

Hi, I'm trying to test this implementation, but unfortunately, I keep getting this error:

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 512, 4, 4]] is at version 2; expected version 1 instead.

Specifically, it is generated at networks.py, line 137.

I've tried different torch versions etc...

Does this also happen to you? I see that in the readme you have torch 1.1.0, but in the paper is reported 1.10.0, which one is the actual version used?

Thanks